The local machine boots using PXE. Clonezilla itself is transferred from a TFTP server as a squashfs and loaded into memory. When that OS boots, it mounts a network share using CIFS that contains the image to be installed. All of the local SATA disks are named sda, sdb, etc. A script determines which SATA disk is the correct one (must be non-rotational, must be a specific size and type), deletes every SCSI device (which includes ATA devices too), then mounts only the chosen disk to make sure it's named sda.

Clonezilla will not allow an image cloned from a device named sda to be written to a device with a different name -- this is why I had to make sure that sda is always the correct SSD.

[–] ulterno@programming.dev 0 points 1 day ago

OIC, so the physically connected storage devices are disconnected in the software and then the correct, required one is re-connected.

The part of what Clonezilla is doing seems like a mis-feature, added to prevent some kind of PICNIC.